NWO.AI 2020 elections forecast: how it works

As the head of machine learning at NWO.AI, I’ve had the opportunity to lead my team to build cutting edge algorithms that parse through several petabytes of human behavioral data to draw insights. I've had a front-row seat to the theatre of the world as it has transformed over the past few months to give rise to a new political, social, and economic order. We have made forecasts on everything from consumer goods to big geopolitical shifts (and turned out to be right). Continuing on our journey, I’m thrilled to announce that NWO.AI has launched a US Presidential Election forecast dashboard. The dashboard, which will be updated twice a week, will show the likelihood (as a percentage) of winning the election, allocated to each candidate on any given day, by our algorithm. For us, this is not just about the horse race, it is about surfacing the underlying meanings behind billions of digital conversations related to each candidate and showcasing how they interact with a candidate’s evolving narrative, as we approach this November.

NWO.AI US Election Forecast

The algorithm primarily takes into account the Impact, Sentiment as well as contextual signals surrounding the two candidates. In the spirit of transparency, we have published a summary of our methodology below. For those who’d like to skip the algorithmic jargon, please follow along this link to proceed to the dashboard.

Background:

Across academic literature, public sentiment derived from social media data has proven uncannily accurate in predicting the outcomes of hard-to-call polls. Knowing this, we have become determined to put to use our massive-scale, data processing capabilities, and advanced analytics techniques, to gauge each of the presidential candidates from the two main parties in the lead up to the 2020 US election, and predict who might win.

Challenges:

An ultra-naive, heuristics-based approach to designing a predictive score would involve simply aggregating the number of positive and negative social media posts mentioning each candidate and deriving the final results proportionally. Digging deeper, however, would reveal that some individuals (or bot controlled accounts) blast hundreds, if not thousands of posts in favor of or opposition to one party or another, spreading highly polar disinformation and disproportionately skewing our results. Also, by analyzing posts in this isolated fashion, we would instantly tear away valuable insight into the dynamics of the network in which the information was distributed, and overlook the psychology of the participating individuals.

Humans are prone to becoming accustomed to herd mentality, where we are heavily conditioned by the opinions of dominating voices around us. In a similar sense, as our opinions solidify, we tend to consume more and more information that reinforces those opinions and disregard contradicting information. It would be presumed then, that a much more meaningful and representative approach to this kind of prediction would involve modeling the interpersonal spread of behaviors throughout the network, and the diffusion of information and sentiment.

Our solution:

Taking these ideas into consideration, we have designed and made use of a 'human-sensor' approach, that profiles the characteristics of individuals within the network, along with the micro-influences occurring between them - rather than simply randomly sampling sentiment from a corpus of posts. That way, we are not only able to understand each individual's political stance, with a certain degree of confidence, but are also able to infer how deterministic or representative their voices are and weight their opinion accordingly.

To be able to conserve the anatomy, physiology, and function of these interactions in the network, and infer directional relationships of influence, we have employed our own adaptation of Googles pivotal algorithm Page Rank, within our own model.

It is not just evolving political opinions towards presidential candidates that can be described and modeled in a socially contagious way, at NWO.AI we apply the same network-wise methodology to monitor and predict all kinds of impending phenomena; from the rise and spread of abstract new ideas to growing geopolitical tension, or even changes in consumer interests and behaviors.

Great! You've successfully subscribed.
Great! Next, complete checkout for full access.
Welcome back! You've successfully signed in.
Success! Your account is fully activated, you now have access to all content.